Applied Image Processing

Frontmatter

1. Design of Industrial Machine Vision Systems

Abstract

There seems to be little disagreement that vision is the most valuable sense that an automaton can possess. The information that it conveys is extremely rich. It can provide absolute and relative position, range, scale, orientation etc., and all this is achieved without the need for physical contact.

G. J. Awcock, R. Thomas

2. Scene Constraints

Abstract

Scene constraint was defined as the first sub-system component in the generic model of a machine vision system presented in figure 1.2. It is a key area in the application of systems engineering principles to the solution of machine vision problems. The designer of machine vision systems should have two principal aims when manipulating scene constraints;

• To maximise the use of prior knowledge of the scene, i.e. by exploiting existing knowledge
• To trivialise the problem of image analysis as far as possible, i.e. by effective imposition of constraints.

G. J. Awcock, R. Thomas

3. Image Acquisition

Abstract

The general aim of the image acquisition sub-system can be summarised as:

‘The transformation of optical image data into an array of numerical data which may be manipulated by a computer, so the overall aim of machine vision may be achieved’.

In order to achieve this aim three major issues must be tackled — these are representation, transduction (or sensing) and digitisation.

G. J. Awcock, R. Thomas

4. Image Preprocessing

Abstract

Image preprocessing seeks to modify and prepare the pixel values of a digitised image to produce a form that is more suitable for subsequent operations within the generic model. There are two major branches of image preprocessing, namely image enhancement and image restoration.

G. J. Awcock, R. Thomas

5. Segmentation

Abstract

The principal objective of the segmentation process is to partition an image into meaningful regions which correspond to part of, or the whole of, objects within the scene. This is done by systematically dividing the whole image up into its constituent areas or regions. If the regions do not correspond directly to a physical object, or object surface, then they should correspond to some area of uniformity as defined by some predetermined assertion, or predicate.

G. J. Awcock, R. Thomas

6. Feature Extraction

Abstract

The feature extraction aspect of image analysis seeks to identify inherent characteristics, or features, of objects found within an image. These characteristics are used to describe the object, or attributes of the object, prior to the subsequent task of classification. Feature extraction operates on two-dimensional image arrays but produces a list of descriptions, or a ‘feature vector’ (note the change in information format indicated on the generic model, section 1.4.1).

G. J. Awcock, R. Thomas

7. Pattern Classification

Abstract

The term pattern classification refers simply to the process whereby an unknown object within an image is identified as belonging to one particular group from among a number of possible object groups. For example, in automatic sorting of integrated circuit amplifier packages there might be three possible types: metal-can, dual-in-line and flat-pack. The unknown object should be classified as being only one of these types.

G. J. Awcock, R. Thomas

8. Image Understanding: Towards Universal Capability

Abstract

A general-purpose machine vision system must be flexible in the sense that it should be able to operate in virtually unconstrained environments containing ill-defined objects which partially occlude one another. Thus the image analysis descriptions of two-dimensional (2-D) relationships must be enriched and extended to include the three-dimensional (3-D) relationships between objects within a real-world scene.

G. J. Awcock, R. Thomas

9. Image Processing Case Histories

Abstract

Computers have many potential uses in the manipulation of images to extract, refine and evaluate information. However, it is extremely taxing of computer power in two respects. Firstly the amount of raw data in an image is very high — a single image of 512 × 512 eight-bit pixels requires 256 kbytes of storage. Secondly it is computationally expensive, with typical non-trivial image processing tasks requiring between 1000 and 10 000 machine cycles per pixel [1]. This translates into either very slow execution times on conventional Von-Neumann architectures or very expensive customised hardware.

G. J. Awcock, R. Thomas

Backmatter